An information-theoretic look at branch-prediction
- Lawrence Livermore National Lab., CA (USA)
- Motorola, Inc., Austin, TX (USA)
Accurate branch-prediction is necessary to utilize deeply pipelined and Very Long Instruction-Word (VLIW) architectures. For a set of program traces we show the upper limits on branch predictability, and hence machine utilization, for important classes of branch-predictors using static (compiletime) and dynamic (runtime) program information. A set of optimal superpredictors'' is derived from these program traces. These optimal predictors compare favorably with other proposed methods of branch-prediction. 3 refs., 5 figs., 12 tabs.
- Research Organization:
- Lawrence Livermore National Lab., CA (USA)
- Sponsoring Organization:
- DOD; DOE/ER; NSF
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 6569918
- Report Number(s):
- UCRL-ID-106077; ON: DE91000499; CNN: CCR-8812843 DAAG29-85-K-0070 N00039-84-; C-0089
- Country of Publication:
- United States
- Language:
- English
Similar Records
An information-theoretic look at branch-prediction
Bulldog: a compiler for VLIW architectures
Toward improved branch prediction through data mining.
Conference
·
Wed Sep 13 00:00:00 EDT 1989
·
OSTI ID:7222477
Bulldog: a compiler for VLIW architectures
Thesis/Dissertation
·
Mon Dec 31 23:00:00 EST 1984
·
OSTI ID:5724953
Toward improved branch prediction through data mining.
Technical Report
·
Tue Sep 01 00:00:00 EDT 2009
·
OSTI ID:993886