Using profiling to reduce branch misprediction costs on a dynamically scheduled processor

ABSTRACT

Modern dynamically scheduled processors use branch prediction hardware to speculatively fetch and execute most likely executed paths in a program. Complex branch predictors have been proposed which attempt to identify these paths accurately such that the hardware can benefit from out-of-order (OOO) execution. Recent studies have shown that inspite of such complex prediction schemes, there still exist many frequently executed branches which are difficult to predict. Predicated execution has been proposed as an alternative technique to eliminate some of these branches in various forms ranging from a restrictive support to a full-blown support. We call the restrictive form of predicated execution as guarded execution.In this paper, we propose a new algorithm which uses profiling and selectively performs if-conversion for architectures with guarded execution support. Branch profiling is used to gather the taken, non-taken and misprediction counts for every branch. This combined with block profiling is used to select paths which suffer from heavy mispredictions and are profitable to if-convert. Effects of three different selection criterias, namely size-based, predictability-based and profiled-based, on net cycle improvements, branch mispredictions and mis-speculated instructions are then studied. We also propose new mechanisms to convert unsafe instructions to safe form to enhance the applicability of the technique. Finally, we explain numerous adjustments that were made to the selection criterias to better reflect the OOO processor behavior.

REFERENCES

• 1: J.C. Dehnert, P.Y.T. Hsu and J. P. Bratt, "Overlapped loop support in the Cydra5," Proceedings of the 16th Annual International Symposium on Computer Microarchitecture, 1989
• 2: Po-Yung Chang , Eric Hao , Yale N. Patt , Pohua P. Chang, Using predicated execution to improve the performance of a dynamically scheduled machine with speculative execution, Proceedings of the IFIP WG10.3 working conference on Parallel architectures and compilation techniques, p.99-108, June 27-29, 1995, Limassol, Cyprus
• 3: P. Y T Hsu , E. S. Davidson, Highly concurrent scalar processing, Proceedings of the 13th annual international symposium on Computer architecture, p.386-395, June 02-05, 1986, Tokyo, Japan
• 4: Scott A. Mahlke , Richard E. Hank , Roger A. Bringmann , John C. Gyllenhaal , David M. Gallagher , Wen-mei W. Hwu, Characterizing the impact of predicated execution on branch prediction, Proceedings of the 27th annual international symposium on Microarchitecture, p.217-227, November 30-December 02, 1994, San Jose, California, United States [doi>10.1145/192724.192755]
• 5: Scott A. Mahlke , David C. Lin , William Y. Chen , Richard E. Hank , Roger A. Bringmann, Effective compiler support for predicated execution using the hyperblock, Proceedings of the 25th annual international symposium on Microarchitecture, p.45-54, December 01-04, 1992, Portland, Oregon, United States
• 6: A. Nicolau and J.A. Fisher, "Measuring the available parallelism for very long instruction word architectures," IEEE- TC, C-33, p 1088-1098 September, 1984
• 7: D. N. Pnevmatikatos , G. S. Sohi, Guarded execution and branch prediction in dynamic ILP processors, Proceedings of the 21ST annual international symposium on Computer architecture, p.120-129, April 18-21, 1994, Chicago, Illinois, United States
• 8: Po-Yung Chang , Eric Hao , Tse-Yu Yeh , Yale Patt, Branch classification: a new mechanism for improving branch predictor performance, Proceedings of the 27th annual international symposium on Microarchitecture, p.22-31, November 30-December 02, 1994, San Jose, California, United States [doi>10.1145/192724.192727]
• 9: B. Ramakrishna Rau , David W. L. Yen , Wei Yen , Ross A. Towie, The Cydra 5 Departmental Supercomputer: Design Philosophies, Decisions, and Trade-Offs, Computer, v.22 n.1, p.12-26, 28-30, 32-35, January 1989 [doi>10.1109/2.19820]
• 10: Gary Scott Tyson, The effects of predicated execution on branch prediction, Proceedings of the 27th annual international symposium on Microarchitecture, p.196-206, November 30-December 02, 1994, San Jose, California, United States [doi>10.1145/192724.192753]
• 11: Tse-Yu Yeh , Yale N. Patt, Two-level adaptive training branch prediction, Proceedings of the 24th annual international symposium on Microarchitecture, p.51-61, September 1991, Albuquerque, New Mexico, Puerto Rico [doi>10.1145/123465.123475]
• 12: Erik Jacobsen , Eric Rotenberg , J. E. Smith, Assigning confidence to conditional branch predictions, Proceedings of the 29th annual ACM/IEEE international symposium on Microarchitecture, p.142-152, December 02-04, 1996, Paris, France
• 13: D. Burger, T.M. Austin and S. Bermett, -"Evaluating Future Microprocessors: The Simplescalar Tools Set," Technical Report TR#1308, University of Wisconsin, 1996
• 14: S. McFarling, "Combining Branch Predictors," WRL Technical Note TR-36, Digital Equipment Corporation, June 1993.
• 15: "WHIRL Intermediate Language Specification," Technical Report, Silicon Graphics Inc., Nov 1999.
• 16: E. Rotenberg , Q. Jacobson , J. Smith, A Study of Control Independence in Superscalar Processors, Proceedings of the 5th International Symposium on High Performance Computer Architecture, p.115, January 09-12, 1999
• 17: John Ruttenberg , G. R. Gao , A. Stoutchinin , W. Lichtenstein, Software pipelining showdown: optimal vs. heuristic methods in a production compiler, Proceedings of the ACM SIGPLAN 1996 conference on Programming language design and implementation, p.1-11, May 21-24, 1996, Philadelphia, Pennsylvania, United States
• 18: S. Mantripragada , S. Jain , J. Dehnert, A New Framework for Integrated Global Local Scheduling, Proceedings of the 1998 International Conference on Parallel Architectures and Compilation Techniques, p.167, October 12-18, 1998
• 19: A. Klauser , T. Austin , D. Grunwald , B. Calder, Dynamic Hammock Predication for Non-Predicated Instruction Set Architectures, Proceedings of the 1998 International Conference on Parallel Architectures and Compilation Techniques, p.278, October 12-18, 1998

INDEX TERMS

Primary Classification:
  I. Computing Methodologies
  I.2 ARTIFICIAL INTELLIGENCE
      I.2.8 Problem Solving, Control Methods, and Search
          Subjects: Scheduling

Additional Classification:
  C. Computer Systems Organization
  C.1 PROCESSOR ARCHITECTURES

  F. Theory of Computation
  F.2 ANALYSIS OF ALGORITHMS AND PROBLEM COMPLEXITY
      F.2.2 Nonnumerical Algorithms and Problems
          Subjects: Sequencing and scheduling

General Terms:
Algorithms, Design, Measurement, Performance, Theory