Fast path-based neural branch prediction

Jiménez, Daniel

doi:10.1109/micro.2003.1253199

Cited by 73 publications

(115 citation statements)

References 23 publications

Supporting

Mentioning

115

Contrasting

Order By: Relevance

“…The Alpha EV6 [19] solves the latency issue by using a simple quick predictor which is backed up by a larger and more accurate predictor. Another approach to solving the latency problem is to use ahead pipelining [11,17,21,26], where the computation required to make a prediction in a particular cycle is started several cycles earlier using the information available in the earlier cycle. However, these schemes do not mitigate the complexity of the hardware itself but instead are themselves very complex.…”

Section: Prior Workmentioning

confidence: 99%

Spotlight - a low complexity highly accurate profile-based branch predictor

Verma

Maderazo

Koppelman

2009

2009 IEEE 28th International Performance Computing and Communications Conference

View full text Add to dashboard Cite

In an effort to achieve the high prediction accuracy needed to attain high instruction throughputs, branch predictors proposed in the literature and used in real systems have become increasingly more complex and larger over time. This is not consistent with the anticipated trend of simpler and more numerous cores in future multi-core processors. We introduce the Spotlight Branch predictor, a novel profile-based predictor which is able to achieve high prediction accuracy despite its simple design. Spotlight achieves high accuracy because complex decisions in the prediction process are made during an OS managed, one-time profile run instead of using complex hardware. We show that Spotlight achieves higher accuracy than Gshare as well as highly accurate and implementable predictors such as YAGS and the Hybrid Bimodal-Gshare predictor. It achieves an average reduction in misprediction rate of 20% over Gshare, 11% over Elastic History Buffer, 14% over Yags and 10% over Hybrid for a hardware budget of 8 kB. Spotlight is also compared to two difficult to implement neural predictors, the Path-based Neural and the Hashed Perceptron. It outperforms the Path-based Neural predictor at all sizes and the Hashed Perceptron at smaller hardware budgets. These results demonstrate that a simple profile-based predictor can achieve many of the benefits of more complex predictors. We also show that a single cycle latency implementation of Spotlight can be achieved without sacrificing accuracy by using an upstream replacement scheme.

show abstract

Section: Prior Workmentioning

confidence: 99%

Spotlight - a low complexity highly accurate profile-based branch predictor

Verma

Maderazo

Koppelman

2009

2009 IEEE 28th International Performance Computing and Communications Conference

View full text Add to dashboard Cite

show abstract

“…In the case of a miss in the TC the prediction is considered wrong, it does not provide any value and a new entry updated with the proper tag and target is added to the respective set, according with the implemented LRU replacement algorithm. The most accurate single-component branch predictors in the literature are neural branch predictors [10], [16] and [17]. Their main advantages consist in the possibility of using longer correlation information at linear cost.…”

Section: Predictors Implemented For Studying Unbiased Branches and Inmentioning

confidence: 99%

“…Unfortunately, the high latency of the perceptron predictor and its impossibility to predict the linearly inseparable branches makes it impractical yet for hardware implementation. In order to reduce the prediction latency, the Fast Path-based Perceptron [10] chooses its weights for generating a prediction according to the current branch's path, rather than according to the branch's PC and history register. The prediction latency is hidden due to the speculative calculation of the perceptron's output.…”

Section: Predictors Implemented For Studying Unbiased Branches and Inmentioning

confidence: 99%

“…Following our aim, first we extend our tool Advanced Branch Prediction trace driven Simulator (ABPS) [8] for analyzing (detecting and predicting) unbiased branches from SPEC JVM98 benchmarks [9], a suite of eight standardized Java applications proposed to better understand where performance bottleneck exist in the Java Virtual Machines and what optimizations are possible. We determine the context length influence on the percentage of unbiased branches from Java applications, the prediction accuracy and the usage degree obtained using a Fast Path-Based Perceptron (FPBP) predictor [10]. Also, we realize a comparison with the behavior of C/C++ applications from unbiased branches perspective.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

The Impact of Java Applications at Microarchitectural Level from Branch Prediction Perspective

Florea¹,

Gellért²,

Vintan³

et al. 2009

INT J COMPUT COMMUN

View full text Add to dashboard Cite

The portability, the object-oriented and distributed programming models, multithreading support and automatic garbage collection are features that make Java very attractive for application developers. The main goal of this paper consists in pointing out the impact of Java applications at microarchitectural level from two perspectives: unbiased branches and indirect jumps/calls, such branches limiting the ceiling of dynamic branch prediction and causing significant performance degradation. Therefore, accurately predicting this kind of branches remains an open problem. The simulation part of the paper mainly refers to determining the context length influence on the percentage of unbiased branches from Java applications, the prediction accuracy and the usage degree obtained using a Fast Path-Based Perceptron predictor. We realize a comparison with C/C++ application behavior from unbiased branches perspective. We also analyze some Java testing programs, built using design patterns or including inheritance, polymorphism, backtracking and recursivity, in order to determine the features of indirect branches, the arity of each indirect jump and the prediction accuracy using the Target Cache predictor.

show abstract

“…However, the ultimate goal-at least when designing a processor-it to build the best predictor. To reach this goal, microarchitects should experiment with using different predictors as prophets and critics; for example, those proposed by Jiménez [13], Seznec [25], and Michaud and Seznec [21].…”

Section: Resultsmentioning

confidence: 99%